≈1.5k tokens on demand. Your agent reads this file only when SKILL.md points to it.
Cast Evolution Engine
Purpose: Define drift detection, severity assessment, change application, confidence decay, and identity protection for EVOLVE.
Contents
- Drift axes
- Severity levels
- Merge rules
- Log and versioning
- Confidence decay
- Batch evolution
- Identity protection
Drift Detection
Four Drift Axes
| Axis |
Detect |
Threshold |
Goals |
Objective shift, changed JTBD, changed priorities |
>=1 significant goal shift |
Pain Points |
New complaints or resolved issues |
>=1 pain point added or removed |
Behavior |
Device mix, frequency, flow, navigation changes |
>=2 behavioral attributes changed |
Segment |
Segment boundaries or dominant user type changes |
Segment boundary changes |
Signal Sources
| Source |
Typical drift signal |
| Trace |
session pattern change, device mix change, drop-off shift |
| Voice |
feedback themes, NPS shift, complaint emergence |
| Pulse |
funnel change, cohort shift, engagement change |
| Field |
new interview findings, revised segments, journey change |
Impact Assessment
| Severity |
Criteria |
Action |
Minor |
1 attribute within 1 axis |
Auto-apply, minor version bump |
Moderate |
2-3 attributes across 1-2 axes |
Auto-apply with notification |
Significant |
4+ attributes across 2+ axes |
Ask first |
Identity |
Role or category would change |
Block and create a new persona |
Typical Change Classes
| Class |
Examples |
| Minor |
device share shift <20%, single behavior refinement, one pain point update |
| Moderate |
tech-level shift, usage-frequency change, multiple pain-point updates |
| Significant |
primary goal change, major behavior shift, segment boundary change, Echo mapping change |
| Identity |
role change, category change, fundamental user-type shift |
Evolution Application
Merge Rules
| Rule |
Meaning |
| Newer wins |
More recent evidence takes precedence |
| Higher confidence wins |
When timestamps are similar, stronger evidence wins |
| Additive preferred |
Add or refine instead of replacing without cause |
| Evidence required |
Every change cites its source |
[inferred] preserved |
Keep inferred markers until real data replaces them |
Section-Level Behavior
| Section |
Evolution behavior |
Profile |
Update attributes except Core Identity |
Quote |
Replace only when new evidence is clearly more representative |
Goals / Frustrations / Behaviors |
Maintain 3; reorder, refine, add, or remove |
Emotion Triggers |
Refine trigger wording and polarity scores |
Context Scenarios |
Add or update scenarios |
JTBD |
Refine, but rarely replace |
Echo Testing Focus |
Update based on fresh friction data |
| Extended sections |
Update attribute-by-attribute with evidence |
Frontmatter Updates
On accepted evolution:
- bump minor
version
- set
status: active
- refresh
updated
- increment
evolution_count
- recalculate
confidence
Evolution Log
Append one row per accepted evolution:
| Field |
Meaning |
Version |
New version |
Date |
ISO date |
Source |
Triggering evidence source |
Changes |
Concise summary |
Confidence Delta |
Signed change in confidence |
Cross-Agent Confidence Feedback
Downstream agents feed findings back to Cast to close the persona lifecycle loop.
| Source Agent |
Signal |
Cast Action |
Confidence Effect |
| Echo |
Walkthrough friction findings for a persona |
FUSE — adjust persona pain points and emotion triggers |
+0.05 if findings align with persona; flag drift if misaligned |
| Echo |
Synthetic persona walkthrough result |
FUSE — validate or challenge proto-persona attributes |
Positive validation: +0.10 toward promotion; Negative: flag for review |
| Echo[demand] |
Demand calibration report ([validated]/[supported]/[hypothesis]) |
FUSE — update persona coverage gaps |
[validated] demands: +0.05; [hypothesis] with no match: flag persona gap |
| Echo[demand] |
Persona coverage gap (demands from segments not in registry) |
CONJURE — consider new persona candidate |
New persona at proto tier (0.30) |
| Trace |
Behavioral validation data |
EVOLVE — existing path (drift detection) |
Per standard drift rules |
| Field |
Interview/survey validation |
FUSE — promote proto → active |
+0.20 (interview) / +0.15 (survey) per validation rules |
Feedback Handoff Format
Downstream agents include the following in their _STEP_COMPLETE or handoff when persona-relevant findings exist:
PERSONA_FEEDBACK:
source_agent: "[Echo | Echo[demand] | Trace | Field]"
persona_id: "[registry ID]"
signal_type: "[validation | drift | gap | calibration]"
findings: "[summary]"
confidence_recommendation: "[+N / -N / review]"
Confidence Decay
| Age since update |
Rule |
<30 days |
No decay |
30-60 days |
-0.05/week |
60-90 days |
-0.10/week |
90+ days |
Freeze and recommend archival review |
Notifications
| Day |
Notification |
Severity |
30 |
Persona stale warning |
Info |
45 |
Confidence decay notice |
Warning |
60 |
Faster decay notice |
Warning |
90 |
Archival review recommendation |
Critical |
Batch Evolution
When running EVOLVE ALL:
- Collect new data.
- Compare each active persona.
- Rank by severity.
- Auto-apply minor/moderate changes.
- Ask before significant changes.
- Return a batch report with
persona / severity / axes / changes / confidence.
Core Identity Protection
Immutable Fields
Identity Change Protocol
- Flag as
Identity.
- Trigger
ON_IDENTITY_CHANGE.
- If approved, create a new persona and archive the old one.
- Cross-reference the old persona in
Source Analysis.
- Update registry coverage and lifecycle state.
